Many platforms have DRAM starting at address 0, which clashes with
barebox placing kernel images there, as barebox does that before
shutting down and turning off the MMU.
The <zero_page.h> header allows temporarily disabling the faulting
zero page and is already used for boot other types of kernel images,
so use it for ELF images as well.
Signed-off-by: Ahmad Fatoum <a.fatoum@xxxxxxxxxxxxxx>
---
common/elf.c | 17 +++++++++++------
1 file changed, 11 insertions(+), 6 deletions(-)
diff --git a/common/elf.c b/common/elf.c
index c9b3d7bb3791..62f793010fad 100644
--- a/common/elf.c
+++ b/common/elf.c
@@ -10,6 +10,7 @@
#include <libfile.h>
#include <memory.h>
#include <unistd.h>
+#include <zero_page.h>
#include <linux/fs.h>
#include <linux/list_sort.h>
@@ -100,7 +101,7 @@ static int elf_section_cmp(void *priv, struct list_head *a, struct list_head *b)
static int load_elf_to_memory(struct elf_image *elf)
{
void *dst;
- int ret, fd = -1;
+ int ret = 0, fd = -1;
u64 p_filesz, p_memsz, p_offset;
struct elf_section *r;
struct list_head *list = &elf->list;
@@ -113,6 +114,8 @@ static int load_elf_to_memory(struct elf_image *elf)
}
}
+ zero_page_access();
+
list_for_each_entry(r, list, list) {
p_offset = elf_phdr_p_offset(elf, r->phdr);
p_filesz = elf_phdr_p_filesz(elf, r->phdr);
@@ -127,14 +130,13 @@ static int load_elf_to_memory(struct elf_image *elf)
if (ret == -1) {
pr_err("lseek at offset 0x%llx failed\n",
p_offset);
- close(fd);
- return ret;
+ goto out;
}
if (read_full(fd, dst, p_filesz) < 0) {
pr_err("could not read elf segment: %m\n");
- close(fd);
- return -errno;
+ ret = -errno;
+ goto out;
}
} else {
memcpy(dst, elf->hdr_buf + p_offset, p_filesz);
@@ -144,9 +146,12 @@ static int load_elf_to_memory(struct elf_image *elf)
memset(dst + p_filesz, 0x00, p_memsz - p_filesz);
}
+out:
+ zero_page_faulting();
+
close(fd);
- return 0;
+ return ret >= 0 ? 0 : ret;
}
static int load_elf_image_segments(struct elf_image *elf)
--
2.39.2
